case-studies/cognida
Finding subrogation cases a rule engine missed
- role
Software/AI Developer, Cognida.ai, Hyderabad, India- when
- Jul 2022 – Aug 2024
- stack
- Python, PySpark, topic modeling, K-means, SQL
Context
Subrogation means recovering a claim payment from the party actually responsible for it. The client screened health insurance claims with a rule engine, which caught the obvious cases and missed the rest.
Approach
- 1PySpark feature engineering over millions of claims
- 2Topic modeling over diagnosis and procedure codes, so related codes group together
- 3K-means clustering on the combined features
- 4A distance-based anomaly score: claims far from their cluster are flagged for review
Result
+8%additional new cases flagged for investigation
Shipped to production alongside the existing rule engine.
Other work there
- ·Retrieval and embedding-based RAG for financial document search, evaluated on retrieval accuracy
- ·Computer vision and OCR classification models for an AI inspection platform, monitored in production
- ·Fine-tuned and quantized an LLM for an insurance-company chatbot to lower serving cost
- ·Compared AWS and Azure for model serving, and helped set up Git, CI/CD and MLOps practices